Multi-Cluster and Hybrid Setup

How the Apollo GraphOS Operator works with subgraphs distributed across multiple clusters or external systems

The Multi-Cluster and Hybrid Setup distributes subgraphs across multiple Kubernetes clusters or external systems while centralizing supergraph deployment. This pattern requires partial: true in SupergraphSchema because not all subgraphs are available in the central cluster.

How It Works

The Operator's Role

  1. Subgraph Discovery: Discovers subgraphs in the central cluster and external endpoints

  2. External Subgraph Management: Subgraphs in remote clusters are published via external endpoints, external systems publish via rover

  3. Hybrid Authoritative: Uses partial: true to compose supergraphs with available subgraphs in the cluster and externally in Studio

  4. Centralized Deployment: Supergraphs are deployed only in the central cluster

Operator Behavior With partial: true

When partial: true is set, the Operator:

  • Discovers and includes Subgraph resources available in the cluster

  • Ignores subgraphs that exist in GraphOS Studio but not in the cluster

  • Sends composition requests to GraphOS Studio when local subgraphs change

  • Updates automatically when subgraphs become available/unavailable

When to Use This Pattern

Use this pattern when:

  • Subgraph implementations are deployed in different clusters

  • Some services run in Kubernetes, others don't

  • You need service isolation for compliance

  • You have multiple Kubernetes clusters

  • You want centralized supergraph management

  • You're migrating from external to Kubernetes

  • You're adopting the Operator from an existing implementation

What's Different About This Pattern

Hybrid Authoritative

  • The Operator ignores subgraphs in GraphOS Studio that aren't defined as Subgraph resources in the cluster and will not remove them

  • GraphOS Studio will compose all subgraphs together, that is, composition is the union of all subgraphs in Studio and the Cluster

Cross-Cluster Communication

  • Subgraphs in remote clusters expose external endpoints

  • The Operator uses these endpoints to inform Routers where to send requests

Centralized Supergraph Management

  • Supergraphs deploy only in the central cluster

  • Teams manage their own subgraph clusters independently

  • Unified supergraph configuration and monitoring

Mixed Subgraph Management

  • Kubernetes subgraphs managed by the Operator

  • External subgraphs published via rover

  • Different update frequencies and workflows
Feedback

Edit on GitHub

Ask Community